Classification of multi-frequency signals with random noise using multilayer neural networks

Frequency analysis capability of multilayer neural networks, trained by backpropagation (BP) algorithm is investigated. Multi-frequency signal classification is considered for this purpose. The number of frequency sets, that is signal groups, is 2/spl sim/5, and the number of frequencies included in a signal group is 3/spl sim/5. The frequencies are alternately located among the signal groups. Through computer simulation, it has been confirmed that the neural network has very high resolution. Classification rates are about 99.5% for trained signals, and 99.0% for untrained signals. The results are compared with conventional methods. Frequency sensitivity and robustness for the random noise are studied. Random noise are added to the multi-frequency signals to investigate how does the network cancel uncorrelated noise among the signals.